Broad band time-varying estimation of event-related synchronization for user-independent configuration of a brain switch

Inter-user variability of brain activity patterns makes it necessary to obtain a user-specific configuration for optimal performance of a brain-computer interface (BCI). Setting up or adjusting a BCI requires a certain amount of data for the computation of spatial filters, selection of suitable features (e.g., frequency bands), and classifier training. Taking advantage of the spatial and spectral characteristics of the post-imagery beta event-related synchronization (ERS, power increase), we propose to use a user-independent configuration encompassing a single Laplacian filter, a broad frequency band (19 to 26Hz), and a novel feature corresponding to the time-varying estimation of the beta ERS. In an offline analysis, we compare i) the use of this novel feature with the traditional logarithmic band power, and ii) the application of a broad frequency band with a user-specific band. After a 10 × 10 cross-validation we found a reduction of the classification accuracy for detection of post-movement beta ERS (4%, 16 participants, p <; 0.05) and post-imagery beta ERS (3%, 9 participants, p > 0.05), when using a general configuration instead of user-specific parameters. This performance reduction was only significant for motor execution. Our results demonstrate that the user-independent configuration proposed here leads to a classification accuracy of post-imagery beta ERS that is not significantly different from the performance achieved when the logarithmic band power of a user-specific frequency band is used.